An advanced image processing technology and an artificial intelligence technology should be combined with each other to develop an image recognition technology. According to the image recognition technology, a situation and an object behavior pattern in an image are recognized through linkage between a low-level process on image information (image segmentation based on location information, depth information and motion information) and a high-level process on the image information (current situation prediction and behavior recognition), in a similar manner as visual functioning of a living thing in which an image is mapped through neurons of a brain so as to be abstraction-processed.
Here, for the low-level process, process techniques such as main feature information extraction, boundary extraction, region division, depth information calculation and motion information calculation with respect to an image are applied. In particular, for the region division, spectrum properties such as light intensity may be used, or shape properties such as eccentricity and density and boundary properties such as curvature and contrast may be used.
According to an image recognition technology according to a related art, a value of pixel intensity difference between adjacent pixels in an image is calculated for the region division based on light intensity. However, such a technology using the value of pixel intensity difference is not able to provide a recognition rate that is robust against external environment change.